You cannot have active read and write lines at the same clock cycle. It is either you read within the command sequence or it is write within the command sequence. The maximum data rate of the esram is 102G/s, assuming 800Mhz clock and 128bit bus.  (It is actually possible for gddr5 to mix reads and writes, since gddr5 has two data clocks but this would be way to complex to explain here).

The clocks of the cpu and the apu can be different. Probably creates a mess when sharing data between cpu and gpu, you are on the safe side with fixed 2:1 ratios, apparently.

So what does reducing the clock speed down from 800Mhz to 750Mhz mean in terms of performance, or is that nothing because the memory clock speed is faster... /me is dumb with these things.

LOL give me that link.

you usually have strong info on this stuff and dont let your personal feelings get in the way of knowledge, could you explain what this means? thanks

Dammit.

Just give us real specs instead to try to make the things better with PR... that will only generate confusion.

I did some more research and the eSRAM needs to run at the clock of the APU... 1:1 ratio.

So we have two possibilities.

1) The GPU is running at 750Mhz
2) The numbers are wrong because they don't add up with 800Mhz.

I disagree

Ps4 is kid buu

X1 is SSj3 Goku

Wiiu is yamcha

If I understand it properly it's saying that during normally running there are spare cycles that can be used to push data around rather than full duplex r/w, hence not quite doubling the speed. It would also explain why the actual speed is considerably lower than the theoretical max because these 'holes' as they call them aren't constant or guaranteed.

Thanks :) But it's just info and not strong info and I don't keep up with recent developments - in my youth I had more time for this ;)

The thing is that usually you have one clock-source (of course you can have several in computing-systems but these are special cases) so everything will be "in sync" with the one clock-source (computing-wise that is). If you have memory which is "faster" (meaning it can operate at higher clocks) you don't have a problem with data from computing units (the ones that generate data that is to be "stored" at memory). I am just saying "memory" but keep in mind that in current systems this includes several "levels" of memory - registers, caches and main memory. I think the abstraction is legit.

Keep in mind this is old info from 90's - nowadays computers don't have fsb's anymore afair when speaking to a pc-spec-specialist recently... ;)

The thing is that if memory doesn't match the speed of the data from cpu you run into problems - it's like letting water run down a hole while the opening is too narrow. The water won't run off fast enough. But thanks to the many levels of memory computers have nowadays the "bad" effect will be not *that* hard but keep in mind that memory is especially intended to serve the computing units. It delivers data to calculate to computing units and will get the stuff that comes from the computing units. If memory is too slow there will be a lot of pipeline-stalls and you have a bottleneck (you always have but they will occur way more often).

There is even more to it like rising and falling edges on clocks but I think this will be too much to explain on a forum like this :)

What did you find on google exactly? Share your info, that's how science works! :)